The purpose of this paper is to present a novel methodology for Defect-Orie
nted (DO) fault sampling, and its implementation in a new extraction tool,
lobs (Layout Observer). The methodology is based on the statistics theory,
and on the application of the concepts of estimation of totals over subpopu
lations and stratified sampling to the fault sampling problem. The proposed
stratified sampling methodology applies to non-equally probable DO faults,
exhibiting a wide range of probabilities of occurrence, and leads to confi
dence intervals similar to the ones obtained with equally probable faults.
ISCAS benchmark circuits are laid out and lobs used to ascertain the result
s, for circuits up to 100,000 MOS transistors, and extracted DO fault lists
of 300,000 faults.